High dimensionality reduction by matrix factorization for systems pharmacology.

Area of Science:

• Computational Biology
• Systems Pharmacology
• Bioinformatics

Background:

• High-dimensional multi-omic data presents challenges in systems pharmacology for understanding drug response and resistance.
• Developing computational methods is crucial for reducing feature dimensionality in diverse biological datasets (in vitro, in vivo, clinical).

Purpose of the Study:

• To propose and evaluate novel feature selection methods based on matrix factorization (MF) for high-dimensional data in systems pharmacology.
• To identify predictive features for phenotype determination and drug response.

Main Methods:

• Utilized matrix factorization (MF) as a core technique for dimensionality reduction.
• Developed three novel feature selection methods based on the mathematical concept of a feature basis.
• Applied these methods alongside three existing MF techniques to analyze eight gene expression datasets.

Main Results:

• The proposed MF methods effectively reduced feature spaces across multiple datasets.
• The techniques successfully identified predictive features related to phenotype determination.
• The three novel methods outperformed existing MF approaches, identifying a 2-gene signature for tyrosine kinase inhibitor response in cell lines.

Conclusions:

• Novel matrix factorization-based feature selection methods are effective for high-dimensional systems pharmacology data.
• These methods can identify key predictive features, including gene signatures, for drug response.
• The developed techniques offer a promising approach for decoding therapeutic responses and resistance mechanisms.